1. Field of the Invention
This invention relates to the field of imaging, and more particularly to the field of pupil sampling for multi-view three-dimensional imaging.
2. Description of the Related Art
One approach to capturing three-dimensional depth information is the use of a pair of two-dimensional recording devices whose relative orientation is known. Much like human or animal optical systems, obtaining depth information from stereo optical systems is a triangulation technique that relies on the difference or disparity between the recorded positions of one or more target points on two imaging sensors. For each target feature, the magnitude of the disparity is directly related to that feature's distance from the imaging system. However, in machine vision systems this approach has shortcomings, such as the high cost of multiple camera/lens systems and the high computational cost of processing large target feature disparities between one sensor and another.
A closely related technology to stereo imaging is triplet imaging, where three separate images are used, rather than two. Triplet imaging is generally used to eliminate processing ambiguity and to provide imaging redundancy, thereby improving accuracy and robustness with minimal addition to computational cost. Most triplet imaging systems consist of three cameras placed equidistant from each other in the form of an equilateral triangle. Stereo systems usually take advantage of rectified camera position that results in disparity only in one direction. In this respect other triplet arrangements (such as an “L” configuration) may yield favorable results. Like stereo-based systems, objects appear displaced in each of the images acquired by these systems with a displacement between cameras proportional to depth. However, unlike stereo systems, the object is displaced in both the horizontal and vertical directions among the three cameras. Once relative camera positions are known, image rectification can reduce the two-dimensional disparities into one-dimensional disparities. To resolve three-dimensional information, the disparity of object features appearing in the three images is determined through triangulation in much the same way as in stereo-based imaging. The three views define all points and edges (lines) within the imaged space. Using more than two sensors assists with working around the so-called “aperture effect” in imaging system, wherein local image disparity can only be determined perpendicular to the texture features providing the signal. This determination requires non-collinear sensors. Having more sensors also assists with occlusion of target surfaces, where one sensor is blocked, but the others can still provide sufficient information, the benefit of which is greater with non-collinear sensors.
In triangulation-based three-dimensional imaging systems (including stereo and triplet imaging systems), a need exists to accurately determine the displacement of object features between acquired images. The processing required to determine this displacement (and thereby allow distinct images to be resolved to a common coordinate system) is commonly referred to as image registration. Many types of image registration processing have been developed, including optical flow (based on the gradients of recorded image intensities), correlation (based on the spatial uniqueness of imaged object feature orientation and intensity), and graph cut (based on minimization of a user defined energy function relating image characteristics).
Thus, a number of mathematical and algorithmic techniques have been developed for resolving optical data into three-dimensional representations of imaged subject matter. However, there remains a need for improved optical trains to acquire data for three-dimensional imaging.